De novo design methods, devices, equipment and storage media for artificial microbial communities
By constructing a genome-scale metabolic network model and improving flow balance analysis, combined with deep learning algorithms, the stability and efficiency issues of microbial community design in large-scale microbial candidate pools were solved, realizing efficient artificial microbial community design and improving production efficiency and application scope.
Patent Information
- Application Number
- CN202410033433.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-01-09
- Publication Date
- 2026-11-10
- Estimated Expiration
- 2044-01-09
AI Technical Summary
Existing technologies struggle to efficiently design artificial microbial communities from large-scale microbial candidate pools, failing to comprehensively consider the metabolism, growth conditions, and interactions among microorganisms, making it difficult to achieve stability and efficiency in microbial community design.
A candidate pool of genome-scale metabolic network models (GEM) was constructed. Biosynthetic pathways were identified through breadth-first search. Combined with improved flow balance analysis and deep learning algorithms, microbial communities with high growth and high metabolic target efficiency were screened out, and their co-occurrence probability was predicted. Stable coexisting artificial microbial communities were designed.
It enables efficient screening of artificial microbial communities that meet preset functions from a large-scale microbial candidate pool, improving production efficiency and yield, and broadening the application scope of microbial factories.
Smart Images

Figure CN117831608B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of microbial community design technology, and in particular to a method, apparatus, device, and storage medium for de novo design of artificial microbial communities. Background Technology
[0002] Designing single microorganisms or microbial communities to degrade or produce desired substances is a hot topic in the field of sustainable biotechnology. Against the backdrop of dwindling fossil resources and global warming, designing microbial communities, rather than using traditional chemical synthesis methods, to accomplish desired functions (such as the synthesis of chemicals based on renewable resources, the synthesis of green fuels, and the degradation of pollutants) is becoming an attractive solution for promoting sustainable biotechnology. The engineering application of single microorganisms has made significant progress in biotechnology, successfully synthesizing a variety of compounds. However, using single microorganisms to synthesize or degrade specific substances presents several challenges, including: 1) Even metabolically engineered single microorganisms still have specific metabolic bottlenecks that limit the complete conversion of substrates into target products; 2) The production of target chemicals often involves the generation of byproducts, reducing product purity and production efficiency; 3) Current bioprocess development relies on existing bioreactors and culture pipelines, which limits innovation and optimization of production processes.
[0003] To address these issues, the design of artificial microbial communities has been proposed as an alternative. These communities can leverage complex interactions among microorganisms to improve the utilization of complex substrates, reduce the demand for genetic load, and minimize the accumulation of byproducts. This has led to progress in both ecology (e.g., soil bioremediation) and biotechnology (e.g., fine chemicals, biopolymers, enzymes, food additives, antimicrobial agents, biofuels, and biohydrogen production).
[0004] However, designing artificial microbial communities naturally faces challenges due to the complexity of intermicrobial interactions. Furthermore, current research on artificial microbial communities often limits itself to using only a small number of seed strains, neglecting to adequately consider other potentially valuable microorganisms. The rational utilization of these potential microorganisms could potentially lead to higher-yield biological processes and the development of entirely new bioproducts, thus broadening the application scope and enhancing the commercial value of microbial factories.
[0005] However, there is currently no method to design microbial communities from a large pool of microbial candidates to achieve efficient biological functions. Summary of the Invention
[0006] This invention provides a method, apparatus, device, and storage medium for de novo design of artificial microbial communities, which at least partially solves the problems existing in related technologies.
[0007] The first aspect of this invention provides a method for de novo design of artificial microbial communities, the method comprising:
[0008] Construct a candidate pool of genome-scale metabolic models (GEMs) for large-scale microorganisms;
[0009] Based on the design objectives of the artificial microbial community, at least one substrate and at least one product are determined; the design objectives of the artificial microbial community include substrate degradation or product production. When the design objective of the artificial microbial community is substrate degradation, the at least one substrate is the target substrate to be degraded, and the at least one product is a product that the target substrate can generate based on a series of metabolic reactions. When the design objective of the artificial microbial community is product production, the at least one product is the target product to be produced, and the at least one substrate is a reactant that can generate the target product based on a series of metabolic reactions.
[0010] For each substrate-to-product reaction pathway: based on the microbial GEM candidate pool, screen for first candidate GEMs that have partial or complete pathways from the substrate to several linkers and finally to the product;
[0011] Based on the steady-state metabolic flux distribution of the first candidate GEM when maximizing metabolic target efficiency and minimizing total flux under the condition of maximizing growth of not less than 99%, a second candidate GEM with high growth and high metabolic target efficiency is screened out. Based on the second candidate GEM, a complete pathway from each substrate to several linkers and finally to the corresponding product is obtained by combination. A combination of multiple second candidate GEMs is determined as a candidate microbial community. By determining the steady-state metabolic flux distribution of the candidate microbial community when maximizing metabolic target efficiency and minimizing total flux under the condition of maximizing growth of not less than 99%, a third candidate GEM with high growth and high metabolic target efficiency is screened out as a candidate microbial community.
[0012] Based on a pre-trained microbial co-occurrence probability prediction model, the co-occurrence probability of the candidate microbial community is predicted. The microbial co-occurrence probability prediction model is trained based on co-occurring communities and mutually exclusive communities.
[0013] Artificial microbial communities are determined based on the predicted co-occurrence probability of candidate microbial communities and the efficiency of maximizing the metabolic target of candidate microbial communities.
[0014] Wherein, when the design goal of the artificial microbial community is substrate degradation, the metabolic goal is target substrate degradation or linker degradation; where the design goal of the artificial microbial community is product production, the metabolic goal is target product production or linker production.
[0015] Optionally, screening first candidate GEMs based on the microbial GEM candidate pool for partial or complete pathways from substrate to several intercalators and finally to the product includes:
[0016] The metabolic reactions and corresponding reactants and products of each GEM are analyzed. All metabolites are treated as nodes and metabolic reactions as hyperedges. The GEM is constructed as a directed hypergraph model. The metabolites include the reactants and products of all metabolic reactions in the GEM.
[0017] Based on the directed hypergraph model, a breadth-first search is performed to determine the GEM containing partial or complete pathways from the substrate to several linkers and finally to the product as the first candidate GEM.
[0018] Optionally, the breadth-first search process includes: using the product as the starting point of the breadth-first search, searching the directed hypergraph model corresponding to each GEM, stopping the search when the search node is a connector or a substrate, and taking that GEM as the first candidate GEM, or
[0019] Using the connector as the starting point for breadth-first search, the directed hypergraph model corresponding to each GEM is searched. When the search node is another connector or the substrate, the search stops and the GEM is taken as the first candidate GEM.
[0020] Optionally, the method further includes:
[0021] Culture medium construction and culture conditions: The culture medium includes LB medium, essential metabolites, and competitive substrates in the culture medium are removed;
[0022] For each first candidate GEM, a single culture of the microorganism is simulated based on the culture medium and culture conditions, and the steady-state metabolic flux distribution is calculated according to the following formula to maximize the target metabolic efficiency and minimize the total flux under the condition of maximizing growth at no less than 99%:
[0023] max v target
[0024]
[0025]
[0026] Among them, v targetThe target metabolic efficiency of the current GEM is represented by z, the total flux is represented by I and J, which represent the set of metabolites and metabolic reactions included in the current GEM, respectively. j This is the flux of reaction j in the current GEM; LB j and UB j These are the lower and upper bounds of metabolic reaction j in the current GEM, respectively; v biomas s represents the current growth of the GEM; This represents the maximum growth, with α being 0.99 to ensure that the growth is not less than 99% of the maximum growth.
[0027] Optionally, the method further includes:
[0028] Culture medium construction and culture conditions: The culture medium includes LB medium, essential metabolites, and competitive substrates in the culture medium are removed;
[0029] Based on the aforementioned culture medium and culture conditions, a co-culture of candidate microbial communities was simulated, and the steady-state metabolic flux distribution of the candidate microbial communities was calculated based on the following formula, which maximizes the metabolic target efficiency and minimizes the total flux under the condition of maximizing growth of at least 99%:
[0030]
[0031]
[0032]
[0033] in, The target metabolic efficiency of the candidate microbial community is represented by z, the total flux is represented by I and J, which represent the sets of metabolites and metabolic reactions, respectively, and N is the set of GEMs contained in the candidate microbial community. It is GEM n The flux of reaction j in the middle; and They are GEM n Lower and upper bounds of the Sino-African exchange reaction j; EXLB j and EXUB j These are the shared lower and upper bounds of the exchange reaction j between microorganisms, respectively; It is GEM n stoichiometric matrix; v total_biomass This represents the total growth of all microorganisms in the candidate microbial community; This represents the maximum growth of all microorganisms in the candidate microbial community, with α being 0.99 to ensure that the growth is not less than 99% of the maximum growth.
[0034] Optionally, the method further includes:
[0035] Collect known co-occurring communities and known mutually exclusive communities as training samples, and label the co-occurrence probability of co-occurring communities as 1 and the co-occurrence probability of mutually exclusive communities as 0. Use the labeled training sample communities to train the preset model.
[0036] During the training process of training a pre-defined model using a training sample population:
[0037] For each training sample community, a feature vector representing its transport response is determined for each microorganism contained therein. The feature vector is a set of 4-dimensional vectors, where each dimension represents the upper and lower bounds of the flux of the metabolic reaction, the charge number of the transported metabolite, and the number of reactions involved. The feature vector contains the transport response profile characteristics of the microorganism.
[0038] For each training sample community, all arrangements of the multiple microorganisms it contains are taken as input. After each arrangement is processed by the preset model, the co-occurrence probability prediction result is output. The average value of the prediction results corresponding to all arrangements is taken as the final prediction result of the preset model for the co-occurrence probability of the training sample community.
[0039] The preset model is trained using the mean squared error between the prediction result of each permutation and the average prediction result as the auxiliary loss function, so as to ensure the invariance of the preset model to the input order as much as possible.
[0040] Optionally, the construction of a candidate pool for genome-scale metabolic network models (GEMs) corresponding to large-scale microorganisms includes:
[0041] Collect known microbial whole genome data or microbial whole proteome data;
[0042] Based on the known microbial whole genome data or microbial whole proteome data, a large-scale GEM candidate pool corresponding to microorganisms is constructed.
[0043] A second aspect of this invention provides an apparatus for de novo design of artificial microbial communities, the apparatus comprising:
[0044] The GEM candidate pool construction module is used to construct GEM candidate pools for large-scale microbial genome-scale metabolic network models.
[0045] The target determination module is used to determine at least one substrate and at least one product based on the design objectives of the artificial microbial community. The design objectives of the artificial microbial community include substrate degradation or product production. When the design objective of the artificial microbial community is substrate degradation, the at least one substrate is the target substrate to be degraded, and the at least one product is a product that the target substrate can generate based on a series of metabolic reactions. When the design objective of the artificial microbial community is product production, the at least one product is the target product to be produced, and the at least one substrate is a reactant that can generate the target product based on a series of metabolic reactions.
[0046] The first screening module is used to screen for each reaction pathway from substrate to corresponding product: based on the microbial GEM candidate pool, a first candidate GEM with a partial or complete pathway from the substrate to several linkers and finally to the product.
[0047] The second screening module is used to screen out second candidate GEMs with high growth and high metabolic target efficiency based on the steady-state metabolic flux distribution of the first candidate GEM under the condition of maximizing the growth rate of not less than 99% and minimizing the total flux. Based on the second candidate GEMs, a complete pathway from each substrate to several linkers and finally to the corresponding product is obtained by combining them. Multiple combinations of second candidate GEMs are identified as candidate microbial communities. By determining the steady-state metabolic flux distribution of the candidate microbial community under the condition of maximizing the growth rate of not less than 99% and minimizing the total flux, a third candidate GEM with high growth and high metabolic target efficiency is screened as a candidate microbial community.
[0048] The co-occurrence probability prediction module is used to predict the co-occurrence probability of the candidate microbial community based on a pre-trained microbial co-occurrence probability prediction model, wherein the microbial co-occurrence probability prediction model is trained based on co-occurring communities and mutually exclusive communities.
[0049] The determination module is used to determine the artificial microbial community based on the predicted co-occurrence probability of the candidate microbial community and the maximum metabolic target efficiency of the candidate microbial community;
[0050] Wherein, when the design goal of the artificial microbial community is substrate degradation, the metabolic goal is target substrate degradation or linker degradation; where the design goal of the artificial microbial community is product production, the metabolic goal is target product production or linker production.
[0051] A third aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described in the first aspect of the present invention.
[0052] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect of the present invention.
[0053] In this embodiment of the invention, a candidate pool of genome-scale metabolic network models (GEMs) corresponding to a large number of microorganisms can be constructed. Utilizing the topology of the metabolic network, a breadth-first search can be used to comprehensively identify biologically significant biosynthetic pathways from the large-scale GEM candidate pool, identifying whether each microorganism possesses a partial or complete pathway from substrate to several linkers and finally to the product. Then, based on an improved multi-step flow balance analysis (FBA) oriented towards high metabolic target efficiency, the steady-state metabolic flux distribution of a single microorganism and microbial community when maximizing the metabolic target efficiency is quantitatively inferred, screening out candidate microbial communities with high growth and high metabolic target efficiency. Finally, a deep learning algorithm is used to predict the co-occurrence probability of the candidate microbial communities, ultimately achieving the de novo design of an artificial microbial community that meets a preset function (including the efficient production or degradation of specified substances) from a large-scale pool of candidate microorganisms. Attached Figure Description
[0054] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0055] Figure 1 A flowchart illustrating the steps of the de novo design method for artificial microbial communities according to an embodiment of the present invention is shown.
[0056] Figure 2 A structural block diagram of an artificial microbial community de novo design device according to an embodiment of the present invention is shown. Detailed Implementation
[0057] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0058] Most current research on microbial community design is typically limited to small-scale candidate pools of microorganisms with well-known genomes or proteomes and manually (or semi-manually) constructed genome-scale metabolic network models. For small-scale candidate pools, preliminary studies have explored how to provide reasonable experimental guidelines for designing microbial communities to achieve predetermined functions, such as CoMiDA, MultiPus, and FLYCOP. However, these methods all have their limitations. For example, graph-based methods like CoMiDA and MultiPus only consider the topology of the microbial metabolic network, but do not consider metabolic flux, culture medium, and microbial composition. They cannot accurately design microbial communities with high growth and high metabolic efficiency, nor can they guarantee the stable coexistence of microbial communities, thus limiting their effectiveness in real-world biological processes. FLYCOP is a flexible method for microbial community modeling and design, but it requires prior knowledge of the relevant microbial community configuration, which is difficult to achieve quickly when dealing with large-scale candidate pools.
[0059] With the development of high-throughput sequencing technology, an increasing number of microbial whole-genome and whole-proteome sequences have been obtained, and corresponding genome-scale metabolic network models have been constructed. Some of these microorganisms have the potential to achieve higher yields and novel biological products. However, screening and precisely designing microbial communities with specific functions from a vast and diverse pool of microbial candidates still faces many challenges. The most significant challenge is that microbial community design involves complex biological and computational problems. As the size of the microbial candidate pool and the number of members in the proposed microbial community increase, the factors that need to be considered grow exponentially, directly leading to a significant increase in computational costs. Furthermore, existing methods often fail to comprehensively consider key factors such as microbial metabolism, growth conditions, and interactions, making it difficult to achieve precise microbial community design. Moreover, the stability of microbial communities and the growth compatibility between different microorganisms are also problems that current technologies struggle to solve. These issues lead to difficulties in the effective translation from theory to experiment, limiting the widespread application of microbial community design.
[0060] Based on this, this invention proposes a de novo design method for artificial microbial communities. This method utilizes the topology of metabolic networks to comprehensively identify biologically significant biosynthetic or degradation pathways from a large-scale GEM candidate pool using a breadth-first search approach. It identifies whether each microorganism possesses a partial or complete pathway from substrate to several linkers and finally to the product. Then, based on an improved flow balance analysis oriented towards high metabolic target efficiency, it quantitatively infers the steady-state metabolic flux distribution of individual microorganisms and microbial communities when maximizing the metabolic target efficiency, and screens out candidate microbial communities with high growth and high metabolic target efficiency. Finally, it uses deep learning algorithms to predict the co-occurrence probability of candidate microbial communities, ultimately achieving the de novo design of artificial microbial communities that meet the preset function (efficient production or degradation of a specified substance) from a large-scale candidate microbial pool.
[0061] The technical solutions provided by the embodiments of the present invention will be described in detail below, with reference to Figure 1 The diagram illustrates a flowchart of a de novo design method for artificial microbial communities according to an embodiment of the present invention. The method may include the following steps:
[0062] S101, construct a candidate pool of GEM (genomic-scale metabolic network) models corresponding to large-scale microorganisms.
[0063] In this embodiment of the invention, each microorganism corresponds to a genome-scale metabolic network model. In this embodiment of the invention, the GEM candidate pool can correspond to a candidate microbial pool.
[0064] S102, based on the design objectives of the artificial microbial community, identify at least one substrate and at least one product.
[0065] The design objectives of the artificial microbial community include substrate degradation or product production. When the design objective of the artificial microbial community is substrate degradation, the at least one substrate is the target substrate to be degraded, and the at least one product is a product that the target substrate can generate based on a series of metabolic reactions. When the design objective of the artificial microbial community is product production, the at least one product is the target product to be produced, and the at least one substrate is a reactant that can generate the target product based on a series of metabolic reactions.
[0066] S103, for each substrate-to-corresponding product reaction pathway: based on the microbial GEM candidate pool, screen for first candidate GEMs that have a partial or complete pathway from the substrate to several linkers and finally to the product.
[0067] S104, based on the steady-state metabolic flux distribution of the first candidate GEM when maximizing metabolic target efficiency and minimizing total flux under the condition of maximizing growth of not less than 99%, a second candidate GEM with high growth and high metabolic target efficiency is screened out; based on the second candidate GEM, a complete pathway from each substrate to several linkers and finally to the corresponding product is obtained by combining them, and a combination of multiple second candidate GEMs is determined as a candidate microbial community; and by determining the steady-state metabolic flux distribution of the candidate microbial community when maximizing metabolic target efficiency and minimizing total flux under the condition of maximizing growth of not less than 99%, a third candidate GEM with high growth and high metabolic target efficiency is screened as a candidate microbial community.
[0068] S105, Based on a pre-trained microbial co-occurrence probability prediction model, predict the co-occurrence probability of the candidate microbial community.
[0069] The microbial co-occurrence probability prediction model is obtained by training on co-occurring communities and mutually exclusive communities.
[0070] S106. Based on the predicted co-occurrence probability of candidate microbial communities and the maximum metabolic target efficiency of candidate microbial communities, the artificial microbial community is determined.
[0071] In this embodiment of the invention, a corresponding GEM pool can be constructed first based on a large-scale microbial pool, and then a breadth-first search algorithm can be used. During the breadth-first search, for each product and substrate determined based on the design objective of the artificial microbial community, the search starting point is the product or the linker. The search stops when another linker or substrate is found, and a first candidate GEM with a partial or complete pathway from the substrate to several linkers and finally to the product is determined.
[0072] In this embodiment of the invention, based on an improved multi-step flow balance analysis oriented towards high metabolic target efficiency, the steady-state metabolic flux distribution for each microorganism is calculated to maximize metabolic target efficiency and minimize the total flux under the condition of maximizing growth at no less than 99%, thus identifying microorganisms with high growth and high metabolic target efficiency as second candidate GEMs. Then, the GEMs corresponding to these high-growth, high-metabolic-target-efficiency microorganisms are linked together using a linker, allowing these microorganisms to jointly complete the process of utilizing substrates to produce products. Specifically, some microorganisms utilize substrates to produce linkers, some microorganisms ingest one linker to produce another, and some microorganisms ingest linkers to produce products. In this embodiment of the invention, multiple microorganisms that can combine to complete this complete pathway from substrate to several linkers and finally to product can be identified to form a candidate microbial community.
[0073] In this embodiment of the invention, based on an improved multi-step flow balance analysis oriented towards high metabolic target efficiency, the steady-state metabolic flux distribution of each candidate microbial community can be determined to maximize metabolic target efficiency and minimize the total flux under the condition of maximizing growth at no less than 99%, thereby selecting communities with high metabolic target efficiency and high growth. Furthermore, the co-occurrence probability of each candidate microbial community is predicted, providing insights into whether the candidate microbial communities can coexist stably; the higher the co-occurrence probability, the higher the probability of stable coexistence. Finally, an artificial microbial community that can coexist stably and exhibits high growth and high metabolic target efficiency is designed.
[0074] In this embodiment of the invention, for any given substrate and product, possible linkers can be automatically selected from a pool of candidate linkers, including small molecule sugars, organic acids, etc. In this embodiment of the invention, considering that crossfeeding is a necessary condition for the cooperation of microbial communities, by designing linkers for crossfeeding, microbial communities with good synergistic relationships can be identified. This embodiment of the invention proposes screening out multiple microorganisms capable of using substrates to produce a series of linkers and ultimately produce products. These microorganisms are considered as multiple functional units, and are progressively connected in a substrate-linker, linker-linker, and finally product manner.
[0075] In this embodiment of the invention, 83 substances were selected as candidate linkers, including two main categories: (1) at least one substance essential for microbial growth; and (2) substances that can be utilized by a large number of microorganisms (more than 10%) in the candidate microbial pool and produced by a large number of microorganisms (more than 10%). Lipopolysaccharides and peptidoglycans (which are cell wall components) and inorganic substances (not considered as effective intermediate products of organic reactions) were excluded.
[0076] In this embodiment of the invention, for each "substrate-to-product" metabolic pathway, all candidate linkers can be traversed to select microorganisms with partial or complete pathways from the substrate to several linkers and finally to the product. Microorganisms with complete pathways from the substrate to several linkers and finally to the product can be selected to form candidate microbial communities. Thus, multiple candidate microbial communities can be obtained based on multiple "substrate-to-product" metabolic pathways.
[0077] In this embodiment of the invention, an artificial microbial community de novo design platform, e-Biota, was developed, which can realize the above steps S101 to S106.
[0078] Specifically, the e-Biota platform includes three new algorithms: the coreBFS algorithm, used to implement step S103; the prodFBA algorithm, used to implement step S104; and the deepCooc algorithm, used to implement step S105. The combined application of these algorithms enables e-Biota to effectively search and design microbial communities with pre-defined functions based on large-scale candidate microbial pools, achieving target substrate degradation and / or target product production, within an acceptable computational cost range.
[0079] In this embodiment of the invention, the inputs to the e-Biota platform typically include target substrates and / or target products; as well as some optional inputs, such as pre-preferred microorganisms, culture media, culture conditions, etc.
[0080] In this embodiment of the invention, based on the constructed microbial GEM candidate pool, coreBFS is used to preliminarily screen candidate microorganisms with partial or complete pathways from substrate to product.
[0081] In this embodiment of the invention, when the design goal of the artificial microbial community is substrate degradation (the target substrate has been determined), the target product can be further specified, that is, the target substrate and the target product are determined. In the breadth-first search, based on all possible pathways from the target substrate to several linkers and finally to the target product, candidate microorganisms with partial or complete pathways from the target substrate to the target product are screened.
[0082] In this embodiment of the invention, when the design goal of the artificial microbial community is substrate degradation (the target substrate has been determined), the target product may not be specified. Instead, all products that the target substrate can generate based on metabolic reactions are considered as potential products. During a breadth-first search, candidate microorganisms with partial or complete pathways from the target substrate to any potential product are screened based on all possible pathways from the target substrate to several linkers and finally to any potential product.
[0083] In this embodiment of the invention, when the design goal of the artificial microbial community is product production (the target product has been determined), the target substrate can be further specified, that is, the target substrate and the target product are determined. In the breadth-first search, based on all possible pathways from the target substrate to several linkers and finally to the target product, candidate microorganisms with partial or complete pathways from the target substrate to the target product are screened.
[0084] In this embodiment of the invention, when the design goal of the artificial microbial community is product production (the target product has been determined), the target substrate may not be specified. Instead, all reactants that can generate the target product based on metabolic reactions are used as potential substrates. During a breadth-first search, candidate microorganisms with partial or complete pathways from any potential substrate to the target product are screened based on all possible pathways from any potential substrate to several linkers and finally to the target product.
[0085] In this embodiment of the invention, optional inputs in the e-Biota platform, such as pre-preferred microorganisms, culture media, culture conditions, etc., can be used to further screen at least one substrate and at least one product for the design objectives based on artificial microbial communities, and to determine the substrates and products for screening the microbial GEM candidate pool.
[0086] In this embodiment of the invention, prodFBA is applied to these candidate microorganisms to further quantitatively infer the steady-state metabolic flux distribution of microorganisms or microbial communities when maximizing metabolic target efficiency and minimizing total flux under the condition of maximizing growth of not less than 99%. Specifically, this includes: firstly, using prodFBA to determine the steady-state metabolic flux distribution of the first candidate GEM when maximizing metabolic target efficiency and minimizing total flux under the condition of maximizing growth of not less than 99%, and screening out the second candidate GEM with high growth and high metabolic target efficiency; then, using prodFBA to determine the candidate microbial community obtained by the combination of candidate microorganisms, calculating the steady-state metabolic flux distribution when maximizing metabolic target efficiency and minimizing total flux under the condition of maximizing growth of not less than 99%, and screening out the candidate microbial community with high growth and high metabolic target efficiency.
[0087] Furthermore, in this embodiment of the invention, the deepCooc method is used to predict the co-occurrence probability of the multiple candidate microbial communities, providing insights into the stable coexistence of microbial communities. Based on the co-occurrence probability and metabolic target efficiency, an artificial microbial community capable of stable coexistence and possessing high growth and high metabolic target efficiency is designed. Thus, in this embodiment of the invention, the de novo design of an artificial microbial community capable of co-occurrence and maximizing preset functions is achieved.
[0088] This invention also provides a method for de novo design of artificial microbial communities. In this embodiment, based on the above embodiments, step S101 includes the following steps:
[0089] S1011, collect known microbial whole genome data or microbial whole proteome data.
[0090] S1012, Based on the known microbial whole genome data or microbial whole proteome data, construct a large-scale GEM candidate pool corresponding to microorganisms.
[0091] In this embodiment of the invention, whole proteome data or whole genome data of microorganisms can be collected from existing public databases (e.g., downloading whole proteome data of microorganisms from the RefSeq database).
[0092] In this embodiment of the invention, a GEM candidate pool corresponding to a microbial community candidate pool can be constructed based on existing GEM construction tools (e.g., the CarveMe algorithm).
[0093] In one exemplary embodiment, after downloading whole proteomic data of 21,514 microorganisms from the RefSeq database, a candidate microbial community GEM library (e-Biota-GEM) is constructed based on the CarveMe algorithm. This database contains 21,514 GEMs, far exceeding existing GEM databases (such as BiGG containing 87 artificially constructed GEMs, AGORA with 818 high-quality GEMs and its upgraded version with 7,302 high-quality GEMs, and CarveMe providing 5,587 GEMs). These 21,514 GEMs cover 37 phyla, 74 classes, 177 orders, 416 families, 1,500 genera, and 5,993 species.
[0094] This invention also provides a method for de novo design of artificial microbial communities. In this embodiment, based on the above embodiments, step S103 includes the following steps:
[0095] S1031, Analyze the metabolic reactions and corresponding reactants and products of each GEM, treat all metabolites as nodes and metabolic reactions as hyperedges, and construct the GEM as a directed hypergraph model. The metabolites include reactants and products.
[0096] In this embodiment of the invention, metabolic reactions may include extracellular reactions and intracellular reactions. Accordingly, metabolites specifically include substances exchanged between the cell and the external environment (uptake and secretion), as well as intracellular metabolites.
[0097] S1032, Based on the directed hypergraph model, perform a breadth-first search to determine the GEM containing a partial or complete pathway from the substrate to several linkers and finally to the product as the first candidate GEM.
[0098] The breadth-first search process includes: using the product as the starting point of the breadth-first search, searching the directed hypergraph model corresponding to each GEM; stopping the search when the search node is a connector or a substrate, and designating that GEM as the first candidate GEM, or...
[0099] Using the connector as the starting point for breadth-first search, the directed hypergraph model corresponding to each GEM is searched. When the search node is another connector or substrate, the search stops and the GEM is taken as the first candidate GEM.
[0100] Optionally, in this embodiment of the invention, during the search process, the hyperedge of the metabolic reaction is simplified into a complete bipartite graph between multiple reactants and products. Nodes with a connectivity greater than a preset threshold are excluded to screen out metabolites with greater biological significance and common byproducts in the metabolic reaction (such as metal ions, oxygen, ATP, and coenzymes) to support the design of artificial microbial communities; the preset threshold can be 20.
[0101] In this embodiment of the invention, it is also proposed that during the search process, the maximum number of reaction steps for each complete pathway be limited to a preset maximum number of steps; common core paths be stored; and a memory search is performed on metabolites that have been searched in GEM to prune redundant paths in order to improve search efficiency.
[0102] In this embodiment of the invention, common core pathways include glycolysis, the tricarboxylic acid cycle, and the pentose phosphate pathway, thereby pre-setting metabolic pathways of related substances, which facilitates the determination of the complete pathway from substrate to several linkers and finally to product.
[0103] In this embodiment of the invention, steps S1031 to S1032 can be implemented based on the coreBFS algorithm. The coreBFS algorithm also pre-stores the microbial GEM candidate pool obtained in step S101, and implements steps S1031 to S1032 based on this microbial GEM candidate pool. CoreBFS is a memory-based breadth-first search algorithm used to quickly identify whether a microbial GEM has a partial or complete pathway from substrate to several linkers and finally to the product.
[0104] In this embodiment of the invention, each GEM in the GEM candidate pool can be saved in SBML format. A program written using Perlv5.26.1 is used to parse the metabolic reactions and corresponding reactants and products of each GEM. Metabolites (including reactants and products) are regarded as nodes and metabolic reactions are regarded as hyperedges, thereby constructing each GEM as a directed hypergraph model.
[0105] In this embodiment of the invention, the hyperedge of the metabolic reaction is simplified into a complete bipartite graph between multiple reactants and products. Byproducts (such as metal ions, oxygen, ATP, and coenzymes) in the reaction are excluded based on a degree greater than 20 in order to capture the connections between key metabolites. Based on the directed hypergraph model after excluding nodes with a degree greater than 20, a memory-based breadth-first search is performed. Thus, in the process of searching for metabolic pathways, the focus is on the substrates and products in the metabolic reaction, while excluding byproducts, thereby improving search efficiency.
[0106] Furthermore, in this embodiment of the invention, the following techniques are applied to improve the efficiency of breadth-first search: 1. Limiting the maximum number of reaction steps for each complete pathway to 30, a threshold sufficient to cover approximately 99% of metabolic reactions. 2. Storing some common core pathways (including glycolysis, the tricarboxylic acid cycle, and the pentose phosphate pathway), which allows for the pre-defined metabolic pathways of related substances. 3. Implementing a memory-based search for metabolites already searched in the GEM, pruning redundant pathways. These improvements not only optimize the search algorithm but also effectively improve search efficiency, making the analysis of metabolic pathways more accurate and efficient.
[0107] This invention also provides a method for de novo design of artificial microbial communities. In this embodiment, step S104 includes S1041 to S1042:
[0108] S1041: Construction of culture medium and culture conditions: The culture medium includes LB medium, essential metabolites, and competitive substrates in the culture medium are removed.
[0109] In this embodiment of the invention, the culture conditions include aerobic and anaerobic conditions, and the culture medium also includes the target substrate.
[0110] S1042: For each first candidate GEM, simulate a single culture of the microorganism based on the culture medium and culture conditions, and calculate the steady-state metabolic flux distribution that maximizes the metabolic target efficiency and minimizes the total flux under conditions of maximizing growth of at least 99% according to the following formula:
[0111] max v target
[0112]
[0113]
[0114] Among them, u target The target metabolic efficiency of the current GEM is represented by z, the total flux is represented by I and J, which represent the set of metabolites and metabolic reactions included in the current GEM, respectively. j This is the flux of reaction j in the current GEM; LB j and UB j These are the lower and upper bounds of metabolic reaction j in the current GEM, respectively; v biomass This indicates the current growth rate of the GEM; This represents the maximum growth, with α being 0.99 to ensure that the growth is not less than 99% of the maximum growth.
[0115] In this embodiment of the invention, for each microorganism, the maximum metabolic target efficiency is calculated based on an improved multi-step flow balance analysis oriented towards high metabolic target efficiency, under the conditions of maximizing growth and minimizing total flux. Microorganisms with high growth and high metabolic target efficiency are identified as second candidate GEMs. Then, these GEMs with high growth and high metabolic target efficiency are linked together through a linker, and these microorganisms jointly complete the process of using substrates to produce products. Among them, some microorganisms use substrates to produce linkers, some microorganisms ingest one type of linker to produce another type of linker, and some microorganisms ingest linkers to produce products.
[0116] Steps S1041 to S1042 can be implemented based on the prodFBA algorithm. The prodFBA algorithm is an improved multi-step flow balance analysis method oriented towards high metabolic target efficiency. It can determine the steady-state metabolic flux distribution of the first candidate GEM when it maximizes metabolic target efficiency and minimizes the total flux under the condition of maximizing growth of at least 99%, and screen out the second candidate GEM with high growth and high metabolic target efficiency.
[0117] This invention also provides a method for de novo design of artificial microbial communities. In this embodiment, step S104 further includes steps S1043 to S1044:
[0118] S1043, Construction of culture medium and culture conditions: The culture medium includes LB medium, essential metabolites, and competitive substrates in the culture medium are removed.
[0119] S1044, based on the culture medium and culture conditions, simulate the co-culture of candidate microbial communities, and calculate and determine the steady-state metabolic flux distribution of candidate microbial communities under the condition of maximizing metabolic target efficiency and minimizing total flux at a growth rate of not less than 99% based on the following formula:
[0120]
[0121]
[0122]
[0123] in, The target metabolic efficiency of the candidate microbial community is represented by z, the total flux is represented by I and J, which represent the sets of metabolites and metabolic reactions, respectively, and N is the set of GEMs contained in the candidate microbial community. It is GEM n The flux of reaction j in the middle; and They are GEM n Lower and upper bounds of the Sino-African exchange reaction j; EXLB jand EXUB j These are the shared lower and upper bounds of the exchange reaction j between microorganisms, respectively; It is GEM n stoichiometric matrix; u total_biomass This represents the total growth of all microorganisms in the candidate microbial community; This represents the maximum growth of all microorganisms in the candidate microbial community, with α being 0.99 to ensure that the growth is not less than 99% of the maximum growth.
[0124] In this embodiment of the invention, steps S1043 to S1044 can also be implemented based on the prodFBA algorithm. The prodFBA algorithm is an improved multi-step flow balance analysis method oriented towards high metabolic target efficiency, which calculates the steady-state metabolic flux distribution of candidate microbial communities when maximizing metabolic target efficiency and minimizing the total flux under the condition of maximizing growth of not less than 99%.
[0125] In specific implementation, the construction of the culture medium and culture conditions in steps S1041 and 1043 are similar, and may specifically include:
[0126] In this embodiment of the invention, the culture medium is constructed starting from LB medium, which is one of the most basic culture media. In order to enable most microorganisms to grow, essential metabolites are added to the culture medium to ensure the growth of most microorganisms, mainly including ions, amino acids, nucleotides, vitamins and various sugars.
[0127] In this embodiment of the invention, essential metabolites refer to substances that, when placed in a culture medium, have a lower limit of exchange reaction flux greater than 0, provided that the biomass is not less than 99% of the maximum biomass. These substances are required to ensure the good growth of the microorganism.
[0128] In this embodiment of the invention, competitive carbon sources in the culture medium are further labeled according to the Biolog phenotypic array, and these competitive carbon sources are removed from the culture medium to reduce the utilization of competitive substrates. Then, the target substrate is added to the culture medium. If the target substrate is not a carbon source, glucose can be added to ensure the presence of a carbon source in the culture medium. Regarding culture conditions, in practical applications, both aerobic and anaerobic conditions can be designed.
[0129] In this embodiment of the invention, after determining the culture medium and culture conditions, the culture of the first candidate GEM can be simulated based on the culture medium and culture conditions, and the steady-state metabolic flux distribution of each first candidate GEM can be calculated based on the formula in step S1042 above, when the metabolic target efficiency is maximized and the total flux is minimized under the condition of maximizing growth of not less than 99%.
[0130] In this embodiment of the invention, after determining the culture medium and culture conditions, the co-culture of microbial communities can be simulated based on the culture medium and culture conditions, and the steady-state metabolic flux distribution of each candidate microbial community can be calculated based on the formula in step S1044 above, when the metabolic target efficiency is maximized and the total flux is minimized under the condition of maximizing growth of not less than 99%.
[0131] In this embodiment of the invention, the prodFBA algorithm was developed to infer the steady-state metabolic flux distribution. This algorithm can simulate the monoculture of microorganisms in various culture media and the co-culture of artificial microbial communities. prodFBA is implemented using Python, mainly using the COBRApy package (v 0.22.0) and the Gurobi solver (v 10.0). Unlike the classic FBA, this algorithm adopts a multi-step strategy guided by high metabolic target efficiency: maximizing substrate degradation efficiency or product production efficiency while ensuring at least 99% maximum production, and minimizing the total flux to reduce alternative pathways, ultimately determining the steady-state metabolic flux distribution.
[0132] In this embodiment of the invention, the prodFBA algorithm can simulate the co-cultivation of a microbial community, treating the entire microbial community as a single microorganism. It isolates the intracellular and periplasmic reactions and metabolites of each microorganism in different compartments, while integrating extracellular reactions and metabolites into a single compartment, representing a shared metabolic environment. The algorithm then calculates the community's production potential, metabolic flux, and interactions under different conditions. These improved strategies enable prodFBA to discover more candidate microbial communities that possess both high potential for metabolic target efficiency and rapid growth, rather than focusing on maximizing growth as exemplified by classical FBA.
[0133] This invention also provides a method for de novo design of artificial microbial communities. In this embodiment, the method further includes:
[0134] Collect known co-occurring communities and known mutually exclusive communities as training samples, and label the co-occurrence probability of co-occurring communities as 1 and the co-occurrence probability of mutually exclusive communities as 0. Use the labeled training sample communities to train the preset model.
[0135] In this embodiment of the invention, during the training process of training a preset model using a training sample community:
[0136] For each training sample community, a feature vector representing its transport response is determined for each microorganism contained therein. The feature vector is a set of 4-dimensional vectors, where each dimension represents the upper and lower bounds of the transport response flux, the charge number of the transported metabolite, and the number of reactions in which the transported metabolite participates in the GEM. The feature vector contains the transport response profile characteristics of the microorganism.
[0137] For each training sample community, all arrangements of the multiple microorganisms it contains are taken as input. After each arrangement is processed by the preset model, the co-occurrence prediction result is output. The average value of the prediction results corresponding to all arrangements is taken as the final prediction result of the preset model for the co-occurrence probability of the training sample community.
[0138] The model is trained using the mean squared error between the prediction result for each permutation and the average prediction result as an auxiliary loss function, in order to ensure the invariance of the model to the input order as much as possible.
[0139] In this embodiment of the invention, 23,324 microbial samples were downloaded from the Earth Microbiome Project (EMP) database. Using the HiOrCo tool, information on co-occurring communities and mutually exclusive communities composed of n microorganisms under natural conditions was obtained from these data, which were designated as co-occurring and mutually exclusive communities, respectively. Specifically: when n=2 (i.e., communities containing two types of microorganisms), 46,005 co-occurring communities and 23,550 mutually exclusive communities were identified; when n=3 (i.e., communities containing three types of microorganisms), 3,812,130 co-occurring communities and 178,815 mutually exclusive communities were identified. In this embodiment of the invention, an independent dataset containing 467 human gut microbiome samples was also obtained. Using the HiOrCo tool, 284 co-occurring communities (n=2) and 11,909 co-occurring communities (n=3) were obtained. All identified co-occurrence and mutually exclusive communities were integrated, and training, validation, and test sets were constructed in a ratio of 18:1:1. The deepCooc model was trained using the training set, parameters were tuned on the validation set to obtain the optimal training model, and the generalization ability of the model was evaluated on the test set.
[0140] In this embodiment of the invention, considering that interactions between microbial communities are mainly achieved through material exchange, for a community containing multiple microorganisms, a feature vector representing its transport response is determined for each microorganism. This is a set of 4-dimensional vectors, where each dimension represents the upper and lower bounds of flux for a specific metabolic reaction, the charge number of transported metabolites, and the number of reactions involved. Therefore, each microorganism is assigned a 4-dimensional vector of length 335, which contains the transport response spectrum characteristics of the microorganism. When the specified microorganism is at the genus level, a certain number (default value 15) of samples from GEMs belonging to that genus can be randomly selected from the GEM candidate pool, vectorized, and the average value is taken as the feature vector of that microorganism.
[0141] In this embodiment of the invention, it is also considered that when processing a community containing multiple microorganisms, the model should not be affected by the input order of the vectors corresponding to the microorganisms. That is, the model is symmetrical with respect to the input order of the microbial community. The vectors corresponding to multiple microorganisms in the microbial community are arbitrarily sorted and the sorting results are input into the model. The model should output the same or similar prediction results.
[0142] Based on this, the embodiments of the present invention employ a full-permutation batch processing data input strategy and an auxiliary loss function. Specifically, deepCooc takes all possible permutations of multiple microorganisms as input, and after processing by the model, outputs a predicted value of the community co-occurrence probability. Then, the average of all permutation predictions is taken as the final prediction result of deepCooc for the co-occurrence probability of that sample. To assist training, in the embodiments of the present invention, the mean squared error between the prediction result of each permutation and the average prediction result is calculated as an auxiliary loss function. Minimizing this auxiliary loss function can ensure that deepCooc is not affected by the input order of the vectors corresponding to the microorganisms. At the same time, during the training process, the main loss function can be determined based on the difference between the predicted value and the true value, and the model is trained with the aim of minimizing the auxiliary loss function and the main loss function.
[0143] In this embodiment of the invention, deepCooc includes multiple single-permutation module structures, each predicting one permutation of the microbial community. The model input is the transport and metabolic response feature vectors of representative microorganisms corresponding to n microorganisms. The feature vector of each microorganism is first extracted through a GEM response network module, and the shared parameters between GEM response network modules of different microorganisms are processed. The GEM response network module consists of four fully connected layers (hidden layer dimensions are 335, 4096, 2048, 1024, and 256, respectively). Each fully connected layer is followed by a dropout layer (dropout rate of 0.5) and a ReLU nonlinear layer.
[0144] Subsequently, a one-dimensional convolutional neural network layer merges the feature representations of n microorganisms processed by the GEM reaction network module and inputs them into the subsequent fully connected module to obtain the prediction result of the co-occurrence probability. The fully connected module consists of three fully connected layers (with hidden layer dimensions of 256×4×n, 1024, 128, and 1, respectively). Each fully connected layer is followed by a dropout layer (with a dropout rate of 0.5) and a ReLU nonlinear layer.
[0145] By employing the methods described above, deepCooc can not only effectively process large datasets and predict the co-occurrence probability of microbial communities, but also maintain invariance to the order of input microorganisms, which is crucial for ensuring the consistency and reliability of model prediction results.
[0146] Based on the same inventive concept, embodiments of the present invention also provide an apparatus for de novo design of artificial microbial communities, referring to... Figure 2 The diagram illustrates a structural block diagram of an artificial microbial community de novo design device according to an embodiment of the present invention. The device includes:
[0147] GEM candidate pool construction module 201 is used to construct a GEM candidate pool for large-scale microbial genome-scale metabolic network models.
[0148] The target determination module 202 is used to determine at least one substrate and at least one product based on the design goals of the artificial microbial community. The design goals of the artificial microbial community include substrate degradation or product production. When the design goal of the artificial microbial community is substrate degradation, the at least one substrate is the target substrate to be degraded, and the at least one product is a product that the target substrate can generate based on a series of metabolic reactions. When the design goal of the artificial microbial community is product production, the at least one product is the target product to be produced, and the at least one substrate is a reactant that can generate the target product based on a series of metabolic reactions.
[0149] The first screening module 203 is used to screen first candidate GEMs with partial or complete pathways from substrate to several linkers and finally to the product based on the microbial GEM candidate pool for each substrate-to-product reaction pathway.
[0150] The second screening module 204 is used to screen out second candidate GEMs with high growth and high metabolic target efficiency based on the first candidate GEMs under the condition of maximizing growth at no less than 99% and minimizing the total flux. Based on the second candidate GEMs, a complete pathway from each substrate to several linkers and finally to the corresponding product is obtained by combining them. A combination of multiple second candidate GEMs is determined as a candidate microbial community. By determining the steady-state metabolic flux distribution of the candidate microbial community under the condition of maximizing growth at no less than 99% and minimizing the total flux, a third candidate GEM with high growth and high metabolic target efficiency is screened as a candidate microbial community.
[0151] The co-occurrence probability prediction module 205 is used to predict the co-occurrence probability of the candidate microbial community based on a pre-trained microbial co-occurrence probability prediction model, wherein the microbial co-occurrence probability prediction model is trained based on co-occurring communities and mutually exclusive communities.
[0152] The determination module 206 is used to determine the artificial microbial community based on the predicted co-occurrence probability of the candidate microbial community and the maximum metabolic target efficiency of the candidate microbial community.
[0153] Wherein, when the design goal of the artificial microbial community is substrate degradation, the metabolic goal is target substrate degradation or linker degradation; where the design goal of the artificial microbial community is product production, the metabolic goal is target product production or linker production.
[0154] Optionally, the first filtering module 203 is specifically used for:
[0155] The metabolic reactions and corresponding reactants and products of each GEM are analyzed. All metabolites are treated as nodes and metabolic reactions as hyperedges. The GEM is constructed as a directed hypergraph model. The metabolites include the reactants and products of all metabolic reactions in the GEM.
[0156] Based on the directed hypergraph model, a breadth-first search is performed to determine the GEM containing partial or complete pathways from the substrate to several linkers and finally to the product as the first candidate GEM.
[0157] Optionally, the breadth-first search process includes: using the product as the starting point of the breadth-first search, searching the directed hypergraph model corresponding to each GEM, stopping the search when the search node is a connector or a substrate, and taking that GEM as the first candidate GEM, or
[0158] Using the connector as the starting point for breadth-first search, the directed hypergraph model corresponding to each GEM is searched. When the search node is another connector or substrate, the search stops and the GEM is taken as the first candidate GEM.
[0159] Optionally, the device further includes:
[0160] The first construction module is used to construct the culture medium and culture conditions: the culture medium includes LB medium, essential metabolites, and removes competing substrates from the culture medium;
[0161] The first calculation module is used to simulate a single culture of the microorganism for each first candidate GEM based on the culture medium and culture conditions, and calculate the steady-state metabolic flux distribution that maximizes the metabolic target efficiency and minimizes the total flux under the condition of maximizing growth of not less than 99% according to the following formula:
[0162] max v target
[0163]
[0164]
[0165] Among them, u target The target metabolic efficiency of the current GEM is represented by z, the total flux is represented by I and J, which represent the set of metabolites and metabolic reactions included in the current GEM, respectively. j This is the flux of reaction j in the current GEM; LB j and UB j These are the lower and upper bounds of metabolic reaction j in the current GEM, respectively; v biomas s represents the current growth of the GEM; This represents the maximum growth, with α being 0.99 to ensure that the growth is not less than 99% of the maximum growth.
[0166] Optionally, the device further includes:
[0167] The second construction module is used to construct the culture medium and culture conditions: the culture medium includes LB medium, essential metabolites, and removes competing substrates from the culture medium;
[0168] The second calculation module is used to simulate the co-culture of candidate microbial communities based on the culture medium and culture conditions, and to calculate and determine the steady-state metabolic flux distribution of the candidate microbial communities under the condition of maximizing the metabolic target efficiency and minimizing the total flux under the condition of maximizing growth of not less than 99% based on the following formula:
[0169]
[0170]
[0171]
[0172] in, The target metabolic efficiency of the candidate microbial community is represented by z, the total flux is represented by I and J, which represent the sets of metabolites and metabolic reactions, respectively, and N is the set of GEMs contained in the candidate microbial community. It is GEM n The flux of reaction j in the middle; and They are GEM n Lower and upper bounds of the Sino-African exchange reaction j; EXLB j and EXUB j These are the shared lower and upper bounds of the exchange reaction j between microorganisms, respectively; It is GEM n stoichiometric matrix; v total_biomass This represents the total growth of all microorganisms in the candidate microbial community; This represents the maximum growth of all microorganisms in the candidate microbial community, with α being 0.99 to ensure that the growth is not less than 99% of the maximum growth.
[0173] Optionally, the device further includes:
[0174] The sample collection module is used to collect known co-occurring communities and known mutually exclusive communities as training samples, and to label the co-occurrence probability of co-occurring communities as 1 and the co-occurrence probability of mutually exclusive communities as 0. The pre-defined model is trained using the labeled training sample communities.
[0175] The training module is used during the training process of training a pre-defined model using a training sample population:
[0176] For each training sample community, a feature vector representing its transport response is determined for each microorganism contained therein. The feature vector is a set of 4-dimensional vectors, where each dimension represents the upper and lower bounds of the flux of the metabolic reaction, the charge number of the transported metabolite, and the number of reactions involved. The feature vector contains the transport response profile characteristics of the microorganism.
[0177] For each training sample community, all arrangements of the multiple microorganisms it contains are taken as input. After each arrangement is processed by the preset model, the co-occurrence probability prediction result is output. The average value of the prediction results corresponding to all arrangements is taken as the final prediction result of the preset model for the co-occurrence probability of the training sample community.
[0178] The preset model is trained using the mean squared error between the prediction result of each permutation and the average prediction result as the auxiliary loss function, so as to ensure the invariance of the preset model to the input order as much as possible.
[0179] Optionally, the GEM candidate pool construction module 201 is specifically used for:
[0180] Collect known microbial whole genome data or microbial whole proteome data;
[0181] Based on the known microbial whole genome data or microbial whole proteome data, a large-scale GEM candidate pool corresponding to microorganisms is constructed.
[0182] Based on the same inventive concept, embodiments of the present invention also provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps in the method described in any of the above embodiments.
[0183] Based on the same inventive concept, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the method described in any of the above embodiments.
[0184] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.
[0185] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, embodiments of the present invention can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, embodiments of the present invention can take the form of computer program products implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0186] Embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (apparatus), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable terminal device, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0187] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable terminal device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0188] These computer program instructions can also be loaded onto a computer or other programmable terminal device to cause a series of operational steps to be performed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal device for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0189] Although preferred embodiments of the present invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present invention.
[0190] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.
[0191] The present invention provides a detailed description of a de novo design method, apparatus, device, and storage medium for artificial microbial communities. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, those skilled in the art will recognize that there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.
Claims
1. A method for de novo design of artificial microbial communities, characterized in that, The method includes: Construct a candidate pool of genome-scale metabolic network models (GEMs) for large-scale microorganisms; Based on the design objectives of the artificial microbial community, at least one substrate and at least one product are determined; the design objectives of the artificial microbial community include substrate degradation or product production. When the design objective of the artificial microbial community is substrate degradation, the at least one substrate is the target substrate to be degraded, and the at least one product is a product that the target substrate can generate based on a series of metabolic reactions. When the design objective of the artificial microbial community is product production, the at least one product is the target product to be produced, and the at least one substrate is a reactant that can generate the target product based on a series of metabolic reactions. For each substrate-to-product reaction pathway: based on the microbial GEM candidate pool, screen for first candidate GEMs that have partial or complete pathways from the substrate to several linkers and finally to the product; Based on the steady-state metabolic flux distribution of the first candidate GEM when maximizing metabolic target efficiency and minimizing total flux under the condition of maximizing growth of not less than 99%, a second candidate GEM with high growth and high metabolic target efficiency is screened out. Based on the second candidate GEM, a complete pathway from each substrate to several linkers and finally to the corresponding product is obtained by combination. A combination of multiple second candidate GEMs is determined as a candidate microbial community. By determining the steady-state metabolic flux distribution of the candidate microbial community when maximizing metabolic target efficiency and minimizing total flux under the condition of maximizing growth of not less than 99%, a third candidate GEM with high growth and high metabolic target efficiency is screened out as a candidate microbial community. Based on a pre-trained microbial co-occurrence probability prediction model, the co-occurrence probability of the candidate microbial community is predicted. The microbial co-occurrence probability prediction model is trained based on co-occurring communities and mutually exclusive communities. Artificial microbial communities are determined based on the predicted co-occurrence probability of candidate microbial communities and the efficiency of maximizing the metabolic target of candidate microbial communities. Wherein, when the design goal of the artificial microbial community is substrate degradation, the metabolic goal is target substrate degradation or linker degradation; where the design goal of the artificial microbial community is product production, the metabolic goal is target product production or linker production.
2. The de novo design method for artificial microbial communities according to claim 1, characterized in that, Screening for first candidate GEMs based on the aforementioned microbial GEM candidate pool, which possess partial or complete pathways from the substrate to several intercalators and ultimately to the product, includes: The metabolic reactions and corresponding reactants and products of each GEM are analyzed. All metabolites are treated as nodes and metabolic reactions as hyperedges. The GEM is constructed as a directed hypergraph model. The metabolites include the reactants and products of all metabolic reactions in the GEM. Based on the directed hypergraph model, a breadth-first search is performed to determine the GEM containing partial or complete pathways from the substrate to several linkers and finally to the product as the first candidate GEM.
3. The de novo design method for artificial microbial communities according to claim 2, characterized in that, The breadth-first search process includes: using the product as the starting point of the breadth-first search, searching the directed hypergraph model corresponding to each GEM; stopping the search when the search node is a connector or a substrate, and designating that GEM as the first candidate GEM, or... Using the connector as the starting point for breadth-first search, the directed hypergraph model corresponding to each GEM is searched. When the search node is another connector or the substrate, the search stops and the GEM is taken as the first candidate GEM.
4. The de novo design method for artificial microbial communities according to claim 1, characterized in that, The method further includes: Culture medium construction and culture conditions: The culture medium includes LB medium, essential metabolites, and competitive substrates in the culture medium are removed; For each first candidate GEM, a single culture of the microorganism is simulated based on the culture medium and culture conditions, and the steady-state metabolic flux distribution is calculated according to the following formula to maximize the target metabolic efficiency and minimize the total flux under the condition of maximizing growth at no less than 99%: max v target Among them, v target The target metabolic efficiency of the current GEM is represented by z, the total flux is represented by I and J, which represent the set of metabolites and metabolic reactions included in the current GEM, respectively. j This is the flux of reaction j in the current GEM; LB h and UB j These are the lower and upper bounds of metabolic reaction j in the current GEM, respectively; v biomass This indicates the current growth rate of the GEM; This represents the maximum growth, with α set to 0.99 to ensure that the growth is not less than 99% of the maximum growth.
5. The de novo design method for artificial microbial communities according to claim 1, characterized in that, The method further includes: Culture medium construction and culture conditions: The culture medium includes LB medium, essential metabolites, and competitive substrates in the culture medium are removed; Based on the aforementioned culture medium and culture conditions, a co-culture of candidate microbial communities was simulated, and the steady-state metabolic flux distribution of the candidate microbial communities was calculated and determined based on the following formula, which maximizes the metabolic target efficiency and minimizes the total flux under the condition of maximizing growth of at least 99%: in, The target metabolic efficiency of the candidate microbial community is represented by z, the total flux is represented by I and J, which represent the sets of metabolites and metabolic reactions, respectively, and N is the set of GEMs contained in the candidate microbial community. It is the flux of reaction j in GEM n; and These are the lower and upper bounds of non-exchange reaction j in GEMn, respectively; EXLB j and EXUB j These are the shared lower and upper bounds of the exchange reaction j between microorganisms, respectively; This is the stoichiometric matrix of GEMn; v total_biomass This represents the total growth of all microorganisms in the candidate microbial community; This represents the maximum growth of all microorganisms in the candidate microbial community, with α being 0.99 to ensure that the growth is not less than 99% of the maximum growth.
6. The de novo design method for artificial microbial communities according to claim 1, characterized in that, The method further includes: Collect known co-occurring communities and known mutually exclusive communities as training samples, and label the co-occurrence probability of co-occurring communities as 1 and the co-occurrence probability of mutually exclusive communities as 0. Use the labeled training sample communities to train the preset model. During the training process of training a pre-defined model using a training sample population: For each training sample community, a feature vector representing its transport response is determined for each microorganism contained therein. The feature vector is a set of 4-dimensional vectors, where each dimension represents the upper and lower bounds of the transport response flux, the charge number of the transport metabolite, and the number of reactions it participates in. The feature vector contains the transport response spectrum characteristics of the microorganism. For each training sample community, all arrangements of the multiple microorganisms it contains are taken as input. After each arrangement is processed by the preset model, the co-occurrence probability prediction result is output. The average value of the prediction results corresponding to all arrangements is taken as the final prediction result of the preset model for the co-occurrence probability of the training sample community. The preset model is trained using the auxiliary loss function that minimizes the mean squared error between the prediction result of each permutation and the average prediction result, so as to ensure the invariance of the preset model to the input order as much as possible.
7. The de novo design method for artificial microbial communities according to claim 1, characterized in that, The construction of the GEM candidate pool for large-scale microbial genome-scale metabolic network models includes: Collect known microbial whole genome data or microbial whole proteome data; Based on the known microbial whole genome data or microbial whole proteome data, a large-scale GEM candidate pool corresponding to microorganisms is constructed.
8. A device for de novo design of artificial microbial communities, characterized in that, The device includes: The GEM candidate pool construction module is used to construct GEM candidate pools for large-scale microbial genome-scale metabolic network models. The target determination module is used to determine at least one substrate and at least one product based on the design objectives of the artificial microbial community. The design objectives of the artificial microbial community include substrate degradation or product production. When the design objective of the artificial microbial community is substrate degradation, the at least one substrate is the target substrate to be degraded, and the at least one product is a product that the target substrate can generate based on a series of metabolic reactions. When the design objective of the artificial microbial community is product production, the at least one product is the target product to be produced, and the at least one substrate is a reactant that can generate the target product based on a series of metabolic reactions. The first screening module is used to screen for each reaction pathway from substrate to corresponding product: based on the microbial GEM candidate pool, a first candidate GEM with a partial or complete pathway from the substrate to several linkers and finally to the product. The second screening module is used to screen out second candidate GEMs with high growth and high metabolic target efficiency based on the steady-state metabolic flux distribution of the first candidate GEM under the condition of maximizing the growth rate of not less than 99% and minimizing the total flux. Based on the second candidate GEMs, a complete pathway from each substrate to several linkers and finally to the corresponding product is obtained by combining them. Multiple combinations of second candidate GEMs are identified as candidate microbial communities. By determining the steady-state metabolic flux distribution of the candidate microbial community under the condition of maximizing the growth rate of not less than 99% and minimizing the total flux, a third candidate GEM with high growth and high metabolic target efficiency is screened as a candidate microbial community. The co-occurrence probability prediction module is used to predict the co-occurrence probability of the candidate microbial community based on a pre-trained microbial co-occurrence probability prediction model, wherein the microbial co-occurrence probability prediction model is trained based on co-occurring communities and mutually exclusive communities. The determination module is used to determine the artificial microbial community based on the predicted co-occurrence probability of the candidate microbial community and the maximum metabolic target efficiency of the candidate microbial community; Wherein, when the design goal of the artificial microbial community is substrate degradation, the metabolic goal is target substrate degradation or linker degradation; where the design goal of the artificial microbial community is product production, the metabolic goal is target product production or linker production.
9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the de novo design method for artificial microbial communities as described in any one of claims 1-7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the de novo design method for artificial microbial communities as described in any one of claims 1-7.
Citation Information
Patent Citations
Method of improvement of organisms using profiling the flux sum of metabolites
CN101460845A
Biomarkers for gastric cancer and uses thereof
CN104204807A
